Fast implementation of evolutionary variable step size algorithm

A new variable step size algorithm, namely the evolutionary variable step size (EVS) algorithm, has been recently proposed by the authors. The EVS algorithm employs the concept of evolution that the step size of the adaptive filter is evolved regularly in a controlled manner and the fittest step size candidate is chosen for the subsequent adaptation. Simulation results show that EVS outperforms the least-mean-square (LMS) algorithm and other variable step size algorithms in terms of convergence time and steady-state misadjustment. However, the required computation during the fitness evaluation period of EVS is rather expensive. In order to balance the computation, a fast implementation of EVS is introduced. The idea of the fast EVS is to select the survivor in the fitness evaluation without updating the weights and distribute the weight update of the survivor throughout the rest of the evolution period. With the fast implementation, the complexity of EVS during the fitness evaluation can be reduced from O(3mN) to O(3N+3 t_e)